Agricultural grain-water-energy-carbon monitoring and early warning method, device, equipment and medium

Abstract

This invention discloses a method, device, equipment, and medium for monitoring and early warning of agricultural food-water-energy-carbon emissions, belonging to the field of smart agriculture technology. The method includes: acquiring production big data of a target area; performing spatial registration based on the production big data and corresponding units in a preset static basic spatial database to construct a performance-environmental condition paired dataset; inputting the performance-environmental condition paired dataset into a causal inference engine to identify direct environmental factors that have a direct causal relationship with crop yield, and eliminating interference factors that only have a correlation relationship, generating a set of sensitive factors for triggering early warning; based on the set of sensitive factors, using a threshold search algorithm to automatically identify the optimal early warning threshold for each key factor affecting crop yield; and monitoring energy crops in the target area based on the set of sensitive factors and the corresponding optimal early warning threshold to obtain monitoring results. This application realizes the monitoring and early warning of land, food, energy, and carbon emissions.

Description

Technical Field

[0001] This invention belongs to the field of smart agriculture and carbon neutrality technology, specifically relating to a method, device, equipment and medium for monitoring and early warning of agricultural food-water-energy-carbon. Background Technology

[0002] Against the backdrop of global efforts to address climate change and promote a green transformation of the energy structure, biomass energy, due to its renewable and carbon-neutral characteristics, has become one of the important ways to replace fossil fuels and reduce greenhouse gas emissions. However, the traditional model of large-scale cultivation of energy crops on arable land inevitably competes with food production for land and water resources, posing a potential threat to national food security.

[0003] Agricultural big data technology has been widely applied in key areas such as real-time agricultural situation monitoring, precise mapping of agricultural resources, and simulation of productivity potential, demonstrating powerful spatiotemporal analysis capabilities. However, existing agricultural monitoring and early warning methods for food, water, energy, and carbon still have shortcomings in their application to agricultural big data. Traditional approaches typically rely on a single land use type or a few soil physicochemical indicators for marginal land identification and suitability evaluation. This model fails to effectively integrate and mine the multi-dimensional nonlinear constraints encompassed in agricultural big data, such as climate fluctuations, topography, ecological red lines, and macroeconomic policies, resulting in low identification accuracy and poor spatial guidance. Furthermore, due to the lack of in-depth collaborative processing of multi-source heterogeneous big data (such as remote sensing images, meteorological observations, and hydrological monitoring), existing methods struggle to automatically generate and quantify the optimal planting scheme that combines environmental and economic benefits while ensuring regional water resource balance and arable land red lines.

[0004] In order to develop and utilize marginal land with low agricultural productivity and low economic value, the common practice is to identify and evaluate the suitability of marginal land based solely on land use type or a few soil indicators. This approach uses a single big data indicator and does not fully consider multi-dimensional constraints such as climate, topography, and policies, resulting in crude identification results and poor spatial guidance. It is difficult to automatically generate and quantify the optimal planting plan that combines environmental and economic benefits while ensuring rigid constraints such as regional water resources and arable land area. Furthermore, when planting on this type of land, due to the crude identification results, it is impossible to achieve dynamic simulation of different future climates, and it is impossible to achieve timely monitoring and early warning of land, food, energy, and carbon emissions. Summary of the Invention

[0005] Based on the above analysis, the embodiments of the present invention aim to provide an agricultural food-water-energy-carbon monitoring and early warning method, device, equipment and medium to achieve timely monitoring and early warning of land, food, energy and carbon emissions.

[0006] On the one hand, embodiments of this application provide an agricultural food-water-energy-carbon monitoring and early warning method, the method comprising: Acquire production big data for the target area, including actual biomass yield, actual meteorological time series, actual management measures records, and actual soil change data; Spatial registration is performed based on the corresponding units in the production big data and the preset static basic spatial database to construct a performance-environmental condition matching dataset. The preset static basic spatial database includes a suitable zoning map for energy crop planting and an initial planting layout scheme. The performance-environmental condition paired dataset is input into the causal inference engine. The causal structure learning algorithm automatically identifies direct environmental factors that have a direct causal effect on crop yield and removes interfering factors that only have a correlation relationship, generating a set of sensitive factors for triggering early warning. Based on the set of sensitive factors, a threshold search algorithm is used to automatically identify the optimal early warning threshold for each key factor affecting crop yield. Based on the set of sensitive factors and the corresponding optimal early warning threshold, energy crops in the target area are monitored to obtain monitoring results, which are used to monitor the food-water-energy-carbon balance in the target area.

[0007] Optionally, acquiring the production big data of the target area includes: By deploying flexible array graphene electrodes in the rhizosphere of crops, soil microcurrent time-series signals are continuously acquired at millisecond-level frequencies. The soil microcurrent time series signal is input into a pre-trained carbon flux fingerprint recognition neural network to identify whether there are characteristic electrical signal waveforms in the time series signal that correspond to the denitrification process, methane oxidation process or carbon fixation metabolism process. When the characteristic electrical signal waveform is identified, the early warning advance amount for an impending abnormal fluctuation in soil microbial metabolic activity is determined based on the occurrence time and amplitude parameters of the characteristic electrical signal waveform. The early warning advance amount is then used as a supplementary field for the time dimension of production big data and simultaneously stored in the performance-environmental condition paired dataset.

[0008] Optionally, after performing spatial registration on corresponding units based on the production big data and the preset static basic spatial database to construct a performance-environment condition paired dataset, the method further includes: For each sample in the performance-environmental condition paired dataset, the corresponding historical early warning records and actual disaster loss records are superimposed according to the spatial location of the sample. Using the historical early warning records and actual disaster loss records as monitoring signals, a counterfactual reasoning algorithm is employed to generate a causal effect difference between "virtual output if no early warning occurred" and "actual output if an early warning occurred" for each sample. The difference in causal effects is used as a weighting factor to perform weighted sampling on the samples in the performance-environment condition paired dataset, generating a causal balance dataset for training the causal inference engine. The aforementioned causal balance dataset was identified as a new performance-environment condition paired dataset.

[0009] Optionally, the step of inputting the performance-environmental condition paired dataset into the causal inference engine, and automatically identifying direct environmental factors with a direct causal effect on crop yield through a causal structure learning algorithm, includes: The performance-environmental condition pairing dataset is input into the initial causal structure learning model, and the set of candidate direct environmental factors and the corresponding initial causal graph structure are output through the constraint-based causal discovery algorithm. For each candidate factor in the set of candidate direct environmental factors, a counterfactual intervention scenario for the candidate factor is constructed in a digital twin environment. The counterfactual intervention scenario includes forcibly setting the candidate factor to a preset intervention value at a preset time point, while keeping the factors other than the candidate factor unchanged. A crop growth model coupled with carbon and nitrogen cycle mechanisms is run in the digital twin environment to simulate the crop yield response trajectory under the counterfactual intervention scenario and generate a virtual yield sequence after the intervention. The virtual yield sequence is compared with the actual yield in the original observation data to calculate the intervention effect value, which represents the degree of difference in yield before and after the intervention. Candidate factors whose intervention effect value is greater than a preset threshold are identified as direct environmental factors with direct causal effects.

[0010] Optionally, the construction of the counterfactual intervention scenario for the candidate factors in the digital twin environment includes: Identify the natural variation patterns of the candidate factors in the time series, and use a time series decomposition algorithm based on Gaussian process regression to split the time series data of the candidate factors into trend terms, seasonal terms, and random disturbance terms; While retaining the trend and seasonal terms, the random disturbance term is replaced with a preset intervention disturbance pattern to generate the intervention sequence corresponding to the candidate factor; The intervention sequence is used as a new value of the candidate factor in the counterfactual scenario and input into the digital twin model to obtain the counterfactual intervention scenario of the candidate factor.

[0011] Optionally, the step of automatically identifying the optimal early warning threshold for each key factor's impact on crop yield based on the set of sensitive factors using a threshold search algorithm includes: For each sensitive factor in the set of sensitive factors, a quantile regression model is constructed with the sensitive factor as the independent variable and multiple conditional quantiles of crop yield as the dependent variable. The quantile regression model estimates the functional relationship between factor values ​​and yield quantiles at multiple quantiles. From the quantile regression results, the conditional quantile function at each quantile is extracted, and the function curve corresponding to each quantile is taken as the output-factor response relationship at the quantile. For a preset risk quantile, search on the function curve of the risk quantile for the factor value that causes the production condition quantile to fall below the preset risk threshold for the first time, and determine the factor value as the optimal early warning threshold.

[0012] Optionally, the monitoring of energy crops in the target area based on the set of sensitive factors and the corresponding optimal early warning threshold, to obtain monitoring results, includes: Obtain a set of sensitive factors and the corresponding causal graph structure. The causal graph structure contains a complete causal path starting from each sensitive factor, passing through at least one intermediate node, and finally reaching the crop yield node. The intermediate nodes include soil process variables, crop physiological variables, and microbial activity variables. On each monitoring unit in the target area, the current value of all factors in the set of sensitive factors is collected in real time, and the current value of each sensitive factor is compared with the corresponding optimal warning threshold to identify the candidate sensitive factors that exceed the threshold for the first time. For each candidate sensitive factor, based on the causal graph structure, determine the causal path from the factor to the output node and all intermediate nodes on the path; Real-time data of each intermediate node is collected sequentially along the causal path, and the real-time data of each intermediate node is compared with the preset intermediate threshold corresponding to the real-time data to determine effective early warning factors. The effective early warning factors include candidate sensitive factors in which the real-time data of all intermediate nodes on the path exceeds the threshold corresponding to the real-time data. Based on the spatial distribution and threshold exceedance of the effective early warning factors, multi-level monitoring results are generated.

[0013] Optionally, after monitoring energy crops in the target area based on the set of sensitive factors and the corresponding optimal early warning threshold to obtain monitoring results, the method further includes: The sensitive factors that trigger the early warning in the monitoring results and their corresponding real-time values ​​are input into the micro-control module deployed in the field to generate an electrical stimulation intervention command. According to the electrical stimulation intervention instructions, a current of a specific frequency is applied to the rhizosphere region of the crop to intervene in the metabolic activity of soil microorganisms; After the intervention ended, intervention-related production big data in the intervention area was collected again, and the intervention-related production big data was compared with the production big data before the intervention to verify the causal relationship and obtain the verification results. If the verification results indicate a weakening of the causal relationship, the early warning effectiveness of the sensitive factor is confirmed, and the weight of the sensitive factor is increased.

[0014] On the other hand, embodiments of this application provide an agricultural food-water-energy-carbon monitoring and early warning device, the device comprising: The acquisition module is used to acquire production big data of the target area, including actual biomass output, actual meteorological time series, actual management measures records, and actual soil change data. The construction module is used to perform spatial registration based on the corresponding units in the production big data and the preset static basic spatial database to construct a performance-environmental condition matching dataset. The preset static basic spatial database includes a suitable zoning map for energy crop planting and an initial planting layout scheme. The input module is used to input the performance-environmental condition pairing dataset into the causal inference engine, which automatically identifies direct environmental factors that have a direct causal effect on crop yield through a causal structure learning algorithm, and removes interfering factors that only have a correlation relationship, generating a set of sensitive factors for triggering early warning. The identification module is used to automatically identify the optimal early warning threshold for each key factor affecting crop yield based on the set of sensitive factors and using a threshold search algorithm. The monitoring module is used to monitor energy crops in the target area based on the set of sensitive factors and the corresponding optimal early warning threshold, and to obtain monitoring results. The monitoring results are used to monitor the food-water-energy-carbon balance in the target area.

[0015] In another aspect, embodiments of this application provide an electronic device, the device comprising: a processor and a memory storing computer program instructions; When the processor executes the computer program instructions, it implements the agricultural food-water-energy-carbon monitoring and early warning method as described in the first aspect.

[0016] In another aspect, embodiments of this application provide a computer storage medium storing computer program instructions, which, when executed by a processor, implement the agricultural food-water-energy-carbon monitoring and early warning method as described in the first aspect.

[0017] In another aspect, embodiments of this application provide a computer program product in which the instructions are executed by the processor of an electronic device, causing the electronic device to perform the agricultural food-water-energy-carbon monitoring and early warning method as described in the first aspect.

[0018] Compared with existing technologies, the agricultural food-water-energy-carbon monitoring and early warning method, apparatus, equipment, and medium of this application can acquire integrated agricultural production big data covering multiple dimensions such as meteorology, soil, hydrology, socio-economics, and policies. Combined with a preset spatial database, a performance-environmental condition paired dataset is constructed. A causal inference engine is used to identify sensitive factors with direct causal effects on crop yield, and the optimal early warning threshold is automatically identified. Therefore, it can achieve monitoring and early warning of food-water-energy-carbon for energy crops in target areas, ensuring high timeliness and accuracy of monitoring and early warning under multiple rigid constraints such as water resource red lines, arable land area, and carbon emission targets. This effectively avoids early warning delays and resource waste caused by limitations in indicators and lack of dynamic simulation capabilities in traditional methods, providing a basis for the development and utilization of marginal land and ensuring the sustainability of agricultural production. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of this specification or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the embodiments of this specification. For those skilled in the art, other drawings can be obtained based on these drawings. Figure 1 This is a flowchart illustrating the agricultural food-water-energy-carbon monitoring and early warning method of this application; Figure 2 A structural block diagram of an agricultural food-water-energy-carbon monitoring and early warning device provided in this application embodiment; Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0021] To facilitate understanding of the embodiments of this application, further explanation and description will be provided below with reference to the accompanying drawings and specific embodiments. These embodiments do not constitute a limitation on the embodiments of this application. In the drawings, the dimensions and relative dimensions of components may be exaggerated for clarity and / or descriptive purposes. When exemplary embodiments can be implemented differently, a specific process sequence may be performed in a different order than that described. For example, two consecutively described processes may be performed substantially simultaneously or in the reverse order of their description. Furthermore, the same reference numerals denote the same components.

[0022] The terminology used herein is for the purpose of describing particular embodiments and is not intended to be limiting. As used herein, unless the context clearly indicates otherwise, the singular forms “a” and “(the)” are also intended to include the plural forms. Furthermore, when the terms “comprising” and / or “including” and variations thereof are used in this specification, it indicates the presence of the stated features, integrals, steps, operations, parts, components, and / or groups thereof, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, parts, components, and / or groups thereof. It should also be noted that, as used herein, the terms “substantially,” “about,” and other similar terms are used as approximate terms rather than as terms of degree, thus explaining the inherent biases in measurements, calculated values, and / or provided values ​​that would be recognized by one of ordinary skill in the art.

[0023] For ease of understanding, the following explains some key terms in this embodiment: Production big data refers to raw information related to crop growth, environmental conditions, and management activities acquired within a target agricultural area through various sensors, manual recording, or remote sensing. It encompasses actual biomass yield, actual meteorological time series, actual management records, and actual soil change data, collectively depicting the overall state of agricultural production.

[0024] A pre-built static basic spatial database is a pre-established database containing geospatial information about the target area. This database typically includes energy crop suitability zoning maps to indicate the suitability of different areas for specific energy crops, as well as initial planting layout plans, providing basic spatial references and planning information for subsequent analysis.

[0025] The performance-environment pairing dataset is a structured collection of data constructed by precisely matching and integrating real-time or historical production big data with spatial units in a pre-defined static underlying spatial database. Each record in this dataset contains crop performance (e.g., yield) at a specific spatial location and its corresponding environmental conditions (e.g., weather, soil, management practices), providing a unified data view for subsequent causal analysis.

[0026] A causal inference engine is a computational system or software module whose core function is to use statistical and machine learning methods to identify causal relationships, rather than merely correlations, between variables in observed data. This engine can process complex agricultural system data to reveal the deep mechanisms affecting crop yields.

[0027] The causal structure learning algorithm is an algorithm that runs within a causal inference engine. Its purpose is to automatically discover the causal graph structure between variables from performance-environment condition paired datasets. This algorithm can distinguish between direct and indirect causal effects and identify factors that have a direct impact on specific outcomes, such as crop yield.

[0028] Direct environmental factors refer to environmental variables that have a direct causal effect on crop yield, identified through causal structure learning algorithms. These factors are the root causes of changes in crop yield, rather than merely phenomena that occur concurrently with yield changes.

[0029] The sensitive factor set is a collection of direct environmental factors identified by the causal inference engine that have a direct causal effect on crop yield. These factors are considered key variables that trigger early warnings because their changes can directly lead to fluctuations in crop yield.

[0030] Threshold search algorithm is a method used to automatically determine the critical values ​​at which key factors affect crop yield from a set of sensitive factors. This algorithm analyzes the relationship between factor values ​​and crop yield to identify warning thresholds where crop yield may change significantly when a factor reaches a certain value.

[0031] The optimal warning threshold is a critical value determined by a threshold search algorithm for each sensitive factor, which can effectively indicate that abnormal fluctuations in crop yield are about to occur. When the real-time value of a sensitive factor exceeds this threshold, the system will trigger the corresponding warning.

[0032] The monitoring results are information generated after real-time or periodic monitoring of energy crops in a target area based on a set of sensitive factors and their optimal early warning thresholds. These results typically include which factors exceeded the thresholds, the extent of the exceedances, and an assessment of the potential impact on crop yield, used to guide subsequent decision-making.

[0033] Food-water-energy-carbon is a comprehensive concept representing four core elements of an agricultural ecosystem: water resource utilization, food production, carbon emissions / sequestration, and energy output (especially biomass energy). Monitoring these four elements aims to achieve resource-efficient utilization, environmental sustainability, and maximized economic benefits in agricultural production.

[0034] To address the problems of existing technologies, this application provides a method, apparatus, equipment, and medium for monitoring and early warning of agricultural food-water-energy-carbon resources. In this application, multi-dimensional production big data is acquired, and a performance-environmental condition paired dataset is constructed using a preset spatial database. A causal inference engine is then used to identify sensitive factors with direct causal effects on crop yield, and the optimal early warning threshold is automatically identified. This enables the monitoring and early warning of food-water-energy-carbon resources for energy crops in target areas, ensuring timeliness and accuracy. It effectively avoids early warning delays and resource waste caused by limitations in indicators and insufficient dynamic simulation capabilities in traditional methods, providing a basis for the development and utilization of marginal land and ensuring the sustainability of agricultural production.

[0035] The agricultural food-water-energy-carbon monitoring and early warning method provided in the embodiments of this application will be introduced first. Figure 1 A schematic flowchart of an agricultural food-water-energy-carbon monitoring and early warning method according to an embodiment of this application is shown. Figure 1 As shown, agricultural food-water-energy-carbon monitoring and early warning methods may include S101-S105: S101, acquire production big data for the target area.

[0036] In this embodiment, production big data may include actual biomass yield, actual meteorological time series, actual management practice records, and actual soil change data. Production big data can be acquired through various methods. For example, actual biomass yield can be obtained through manual weighing, remote sensing image analysis, or estimation using empirical models. Actual meteorological time series can be provided by regional meteorological stations or periodically collected using conventional meteorological sensors deployed in the fields. Actual management practice records can be manually recorded by farmers, such as fertilizer application rates, irrigation times, and pest and disease control measures. Actual soil change data can be collected through periodic sampling and analysis of soil physicochemical properties, such as soil pH, organic matter content, and nitrogen, phosphorus, and potassium content, or through general-purpose soil sensors. After collection, this data can be stored on a local server or in a cloud database for subsequent processing.

[0037] S102, based on the corresponding units in the production big data and the preset static basic spatial database, performs spatial registration to construct a performance-environmental condition pairing dataset.

[0038] In this embodiment, after acquiring the production big data, it needs to be spatially registered with a pre-set static basic spatial database to construct a performance-environmental condition matching dataset. Spatial registration can be achieved using Geographic Information System (GIS) software, matching the geographic coordinates in the production big data with spatial units (e.g., plots, grid units) in the pre-set static basic spatial database. The pre-set static basic spatial database may contain a zoning map of energy crop planting suitability, which can be divided based on large-scale factors such as topography, climate, and soil type, as well as an initial planting layout plan, which can be derived from expert experience or historical data analysis. Through this registration, the production big data (such as yield, weather, soil, etc.) of each spatial unit can be associated with the static environmental information (such as suitability level, initial layout) of that unit, forming a dataset containing crop performance and corresponding environmental conditions.

[0039] The pre-set static basic spatial database includes a suitable zoning map for energy crop planting and an initial planting layout plan; S103. Input the performance-environmental condition paired dataset into the causal inference engine. The causal structure learning algorithm automatically identifies direct environmental factors that have a direct causal effect on crop yield and removes interfering factors that only have a correlation relationship, generating a set of sensitive factors for triggering early warning.

[0040] In this embodiment, the constructed performance-environmental condition paired dataset is input into a causal inference engine. Within this engine, a causal structure learning algorithm is used to analyze the data to identify environmental factors with a direct causal effect on crop yield. For example, statistical methods based on Bayesian networks or structural equation modeling can be used to construct a causal graph by analyzing the conditional independence or path coefficients between variables. This algorithm aims to distinguish confounding factors that are only statistically correlated but have no causal relationship. For example, ice cream sales and the number of drownings may be correlated, but they are not directly causally related; rather, they are influenced by the common factor of summer temperature. In this way, the direct environmental factors that truly affect crop yield can be screened from numerous environments, and these factors can be aggregated into a sensitive factor set as the basis for subsequent early warning.

[0041] S104, based on a set of sensitive factors, uses a threshold search algorithm to automatically identify the optimal early warning threshold for each key factor affecting crop yield.

[0042] In this embodiment, after determining the set of sensitive factors, a warning threshold needs to be set for each sensitive factor. The threshold search algorithm can employ various statistical or machine learning methods. For example, it can be determined based on historical data by analyzing the statistical distribution relationship between sensitive factor values ​​and crop yield decline. One approach is to set the warning threshold when crop yield begins to decline significantly when a sensitive factor (such as rainfall) falls below a certain historical average or safety lower limit. Alternatively, classification algorithms such as decision trees and support vector machines can be used, employing "normal yield" and "abnormal yield" from historical data as labels to train a model to identify the range of sensitive factor values ​​that cause yield anomalies, thereby determining the optimal warning threshold. These thresholds aim to indicate that crop yield may face risks when a sensitive factor reaches or exceeds a certain value.

[0043] S105, based on the set of sensitive factors and the corresponding optimal early warning threshold, monitors energy crops in the target area and obtains monitoring results.

[0044] In this embodiment, the monitoring results are used to monitor the food-water-energy-carbon balance of the target area. After determining the set of sensitive factors and the optimal warning threshold, the system continuously monitors energy crops in the target area. This includes real-time or near-real-time acquisition of the current values ​​of sensitive factors (such as soil moisture, temperature, and light intensity). The acquired real-time data is compared with the pre-set optimal warning threshold. Once the current value of a sensitive factor exceeds its corresponding threshold, the system generates a monitoring result indicating that the factor may have a negative impact on crop yield. For example, if soil moisture is below the warning threshold, a drought warning may be triggered. These monitoring results can be presented to managers in the form of reports, charts, or alarms for comprehensive monitoring and management of water resource utilization, food yield potential, carbon emissions / carbon sequestration capacity, and biomass energy production in the target area.

[0045] In this embodiment, multi-dimensional production big data is acquired and a performance-environmental condition paired dataset is constructed using a preset spatial database. A causal inference engine is then used to identify sensitive factors with direct causal effects on crop yield and to automatically identify the optimal early warning threshold. This enables the monitoring and early warning of energy crops' food-water-energy-carbon balance in target areas, ensuring timeliness and accuracy. It effectively avoids early warning delays and resource waste caused by limitations in indicators and insufficient dynamic simulation capabilities in traditional methods, providing a basis for the development and utilization of marginal land and ensuring the sustainability of agricultural production.

[0046] In some other embodiments, S101 may include: By deploying flexible array graphene electrodes in the rhizosphere of crops, soil microcurrent time-series signals are continuously acquired at millisecond-level frequencies. The soil microcurrent time series signal is input into a pre-trained carbon flux fingerprint recognition neural network to identify whether there are characteristic electrical signal waveforms in the time series signal that correspond to the denitrification process, methane oxidation process or carbon fixation metabolism process. When a characteristic electrical signal waveform is identified, the early warning lead time for an impending abnormal fluctuation in soil microbial metabolic activity is determined based on the occurrence time and amplitude parameters of the characteristic electrical signal waveform. This early warning lead time is then used as a supplementary field to the time dimension of production big data and simultaneously stored in the performance-environmental condition paired dataset.

[0047] In this embodiment, the flexible array graphene electrode is a sensor with good flexibility and biocompatibility, capable of closely adhering to the rhizosphere environment of organisms and minimizing disturbance to soil structure. Its array design allows for simultaneous data acquisition at different spatial points, providing more comprehensive spatial information. The high conductivity and large specific surface area of ​​graphene material make it highly sensitive to weak electrochemical signals, enabling continuous capture of extremely weak current changes in the soil at millisecond frequencies, ensuring the temporal resolution and precision of the data. These acquired soil microcurrent time-series signals are electrochemical responses generated by microbial metabolic activities, ion migration, and other biogeochemical processes in the soil. Through high-frequency acquisition, continuous time-series data can be formed, reflecting the dynamic changes in the soil microenvironment.

[0048] Subsequently, the soil microcurrent time-series signal was input into a pre-trained carbon flux fingerprinting neural network. This neural network is an intelligent model trained on a large amount of historical data, capable of automatically identifying unique electrical signal patterns, or "fingerprints," associated with specific microbial metabolic processes from complex soil microcurrent time-series signals. These specific microbial metabolic processes include denitrification, methane oxidation, or carbon fixation metabolism, which are key carbon and nitrogen cycle processes in the soil, directly affecting soil fertility, greenhouse gas emissions, and carbon sequestration capacity. For example, denitrification leads to nitrogen loss, methane oxidation helps reduce methane emissions, and carbon fixation metabolism increases soil organic carbon. Characteristic electrical signal waveforms refer to electrical signal patterns in the soil microcurrent time-series signals that correspond to the aforementioned specific microbial metabolic processes, possessing unique shapes, frequencies, amplitudes, or durations. These waveforms are a direct electrical manifestation of microbial activity.

[0049] When a characteristic electrical signal waveform is identified, the system determines the advance warning amount for an impending abnormal fluctuation in soil microbial metabolic activity based on the waveform's occurrence time and amplitude parameters. For example, if a waveform is detected to show a gradually increasing trend in intensity and frequency, the system can predict that related microbial activity will reach an abnormal level at a future point in time, thus calculating an advance warning amount. This advance warning amount is used as a supplementary field to the time dimension of the production big data, stored along with actual biomass yield, meteorological data, etc., and synchronously stored in the performance-environmental condition paired dataset. The performance-environmental condition paired dataset is a comprehensive dataset that includes crop performance (such as yield) and various environmental conditions (such as soil, meteorology, and management practices). By adding the advance warning amount as a supplementary field to the time dimension, this dataset can more comprehensively reflect the dynamic and predictive relationship between environmental conditions and crop performance.

[0050] The deployment of flexible array graphene electrodes enables millisecond-level continuous acquisition of microcurrents in crop rhizosphere soil, thereby capturing microscopic electrical signals related to microbial metabolic activities that are difficult to detect using traditional methods. A pre-trained carbon flux fingerprinting neural network can accurately identify characteristic waveforms in these signals corresponding to key processes such as denitrification, methane oxidation, or carbon fixation metabolism. This allows for early warning based on the waveform's occurrence time and amplitude parameters, determining the lead time before these microbial activities are about to fluctuate abnormally. This early warning lead time is used as a supplementary field to the time dimension of production big data and simultaneously stored in the performance-environmental condition paired dataset, greatly enriching the dataset's predictive capabilities. This enables the subsequent causal inference engine to incorporate more forward-looking microbial activity information when identifying sensitive factors with direct causal effects on crop yield, thus identifying potential risk factors earlier and more accurately, and providing more timely and refined decision support for monitoring and early warning of food-water-energy-carbon resources, significantly improving the accuracy and timeliness of early warnings. In some other embodiments, after S102, the method may further include: For each sample in the performance-environmental condition paired dataset, overlay the corresponding historical early warning records and actual disaster loss records according to the spatial location of the sample. Using historical early warning records and actual disaster loss records as monitoring signals, a counterfactual reasoning algorithm is used to generate a causal effect difference between "virtual output if no early warning occurred" and "actual output if an early warning occurred" for each sample. The difference in causal effects is used as a weighting factor to weight the samples in the performance-environment condition paired dataset, generating a causal balanced dataset for training the causal inference engine. The causal balance dataset was identified as a new performance-environment condition pairing dataset.

[0051] Specifically, for each sample in the performance-environmental condition pairing dataset, corresponding historical early warning records and actual disaster loss records are overlaid according to the sample's spatial location, aiming to enrich the contextual information of each data sample. Each sample in the performance-environmental condition pairing dataset represents the crop growth status and environmental conditions at a specific time and spatial location. By overlaying historical early warning records and actual disaster loss records, information such as whether an early warning was triggered in the past and whether actual disaster losses were suffered can be associated with each sample. This can be achieved through Geographic Information System (GIS) technology, which performs spatial and temporal matching between sample data with spatial coordinates and historical early warning records (e.g., time, type, and region of the early warning) and actual disaster loss records (e.g., time, type, scope of impact, and degree of loss of the disaster) that also have spatial and temporal stamps. For example, for a data sample of a specific plot in a certain growing season, it is possible to query whether the plot has experienced any early warning events or disaster events in the same growing season or related time period in the past, and add the detailed information of these events as new attribute fields to the sample. These records can come from historical monitoring data, disaster reports, agricultural statistics, etc.

[0052] Based on this, using historical early warning records and actual disaster loss records as supervisory signals, a counterfactual inference algorithm is employed to generate a causal effect difference between "virtual yield if no warning had occurred" and "actual yield if a warning had occurred" for each sample. This step is crucial, aiming to quantify the true causal impact of early warning events on crop yield. Traditional statistical correlation analysis cannot distinguish between causal relationships and confounding factors. The counterfactual inference algorithm can simulate a virtual scenario of "what would the yield be if the warning had not occurred" and compare it with the actual yield when the warning occurred, thereby calculating the causal effect of the warning. Historical early warning records and actual disaster loss records serve as supervisory signals, guiding the counterfactual inference model to learn the true causal mechanism between early warnings and yield. Counterfactual inference algorithms can employ various methods, such as propensity score matching (PSM), difference-in-differences (DiD), structural causal models (SCM), or machine learning-based counterfactual prediction models (such as CausalForests, Dragonnet, etc.). For example, for a sample that experienced a warning, a control sample that is highly similar to the sample in all observable covariates (environmental conditions, management practices, etc.) but did not experience a warning can be found. Then, the actual yield of the control sample is used as a "virtual yield if no warning had occurred," and compared with the actual yield of the sample that experienced a warning to calculate the causal effect difference. This causal effect difference quantifies the degree of impact of the warning event (or the intervention it represents) on crop yield.

[0053] Furthermore, the causal effect difference is used as a weighting factor to perform weighted sampling on samples in the performance-environment condition paired dataset, generating a causal balanced dataset for training the causal inference engine. This step aims to correct the bias in the original dataset by adjusting sample weights, enabling the causal inference engine to learn the true causal relationships more accurately during training. A larger causal effect difference indicates a more significant impact of the warning event on the output of that sample; therefore, it should be given a higher weight during training to ensure the model can fully learn these key causal mechanisms. Weighted sampling can be implemented in several ways. For example, samples in the original dataset can be drawn with replacement based on the magnitude of the causal effect difference, with samples having a larger causal effect difference being drawn more frequently and thus appearing more often in the new dataset. Alternatively, each sample can be assigned a weight value proportional to the absolute value of the causal effect difference. During training, the model will adjust the loss function or gradient updates based on these weights, giving greater emphasis to fitting high-weight samples. Ultimately, the causal balance dataset was identified as the new performance-environment condition pairing dataset, ensuring that subsequent causal inference engine training and sensitive factor identification were based on this more accurate and causally representative dataset.

[0054] This application's embodiments introduce historical early warning and disaster loss records, and utilize counterfactual reasoning algorithms to quantify the true causal effect of early warning events on crop yield, thereby generating a causal effect difference. This difference is used as a weighting factor to weight the original dataset, constructing a causal balance dataset. This process effectively corrects potential sample bias and confounding factors in the original observation data, enabling the causal inference engine to be trained based on more causally representative data. Therefore, this application can significantly improve the accuracy and robustness of the causal inference engine in identifying direct environmental factors with a direct causal effect on crop yield, avoiding erroneous causal identification caused by historical data bias or sample imbalance. This makes the subsequently generated sensitive factor set more reliable, the early warning threshold more accurate, and ultimately improves the effectiveness and reliability of agricultural food-water-energy-carbon monitoring and early warning.

[0055] In some other embodiments, S103 may include: The performance-environmental condition pairing dataset is input into the initial causal structure learning model, and the set of candidate direct environmental factors and the corresponding initial causal graph structure are output through the constraint-based causal discovery algorithm. For each candidate factor in the set of candidate direct environmental factors, a counterfactual intervention scenario for the candidate factor is constructed in the digital twin environment. The counterfactual intervention scenario includes forcibly setting the candidate factor to a preset intervention value at a preset time point, while keeping the factors other than the initial candidate factor unchanged. Running a crop growth model coupled with carbon and nitrogen cycle mechanisms in a digital twin environment simulates the crop yield response trajectory under counterfactual intervention scenarios, generating a virtual yield sequence after intervention; The virtual yield sequence is compared with the actual yield in the original observation data to calculate the intervention effect value, which represents the degree of difference in yield before and after the intervention. Candidate factors whose intervention effect value is greater than a preset threshold are identified as direct environmental factors with direct causal effects.

[0056] Specifically, firstly, the performance-environmental condition pairing dataset is input into the initial causal structure learning model. A constraint-based causal discovery algorithm then outputs a set of candidate direct environmental factors and the corresponding initial causal graph structure. This step aims to initially screen candidate factors that may have a causal relationship with crop yield from a large number of environmental factors and construct a preliminary causal relationship network. The performance-environmental condition pairing dataset contains big data on production in the target area, such as actual biomass yield, actual meteorological time series, actual management records, and actual soil change data, as well as information such as energy crop planting suitability zoning maps and initial planting layout schemes from a pre-set static basic spatial database. The initial causal structure learning model can be implemented using various algorithms, such as the PC algorithm, FCI algorithm, or GES algorithm. These algorithms infer the causal structure between variables by analyzing the conditional independence in the data.

[0057] In some embodiments, to quantify the explanatory power of sensitivity factors on yield spatial differentiation and to determine the impact of sensitivity factors on energy crop yield, the sensitivity factors can be determined using the following formula: Where q represents the explanatory power of environmental factors on the spatial distribution of output; h represents the number of spatial strata or zones of environmental impact factors; N h N and N represent the number of samples in layer h and the total number of samples in the study area, respectively; and These represent the variance of production within layer h and the total variance of production in the study area, respectively.

[0058] Secondly, for each candidate factor in the set of candidate direct environmental factors, a counterfactual intervention scenario is constructed in the digital twin environment. The counterfactual intervention scenario involves forcibly setting the candidate factor to a preset intervention value at a predetermined time point, while keeping other factors unchanged. The digital twin environment is a virtual high-fidelity model of the agricultural ecosystem of the target area, capable of simulating complex interactions in the real world. The purpose of constructing the counterfactual intervention scenario is to simulate the effect of precise intervention on a single candidate factor in a controlled virtual environment, thereby isolating its independent impact on crop yield. For example, if "soil moisture content" is a candidate factor, it can be set in the digital twin environment to forcibly maintain soil moisture content at a specific low value during critical crop growth periods, while other environmental factors (such as temperature, light, fertilizer application, etc.) evolve according to their original observed or predicted trajectories.

[0059] Building upon this, a crop growth model coupled with carbon and nitrogen cycling mechanisms is run in a digital twin environment to simulate the crop's yield response trajectory under counterfactual intervention scenarios, generating a virtual yield sequence after the intervention. The crop growth model coupled with carbon and nitrogen cycling mechanisms is a complex biogeochemical model capable of simulating crop growth and development under different environmental conditions, including photosynthesis, respiration, nutrient uptake, biomass accumulation, and the cycling of carbon and nitrogen in the soil-plant system. By using counterfactual intervention scenarios as input, the model can predict the crop's growth status and final yield under specific intervention conditions, thereby generating a virtual yield time series or final yield value.

[0060] In other embodiments, since energy crops are affected by climate when emitting carbon, which can influence the determination of sensitive factors, downscaling calculations can be performed on climatic factors such as temperature and precipitation. The specific formulas are as follows: Among them, T fut,y T represents the predicted daily temperature for year y; obs,hist Represents historical baseline weather station observations or high-precision gridded measured data; T gcm,fut,y This represents the simulated daily temperature values ​​for the future year y, output by the global model. This represents the daily average simulated value of the GCM over the corresponding historical baseline period.

[0061] Taking into account the impact of climate on direct environmental factors, extreme weather events can be used as an example. For extreme weather events, based on T fut,y Calculate the number of days of high-temperature heatwave (HW).

[0062] In the formula, T fut,y,dT represents the simulated daily temperature value on day d of year y; D is the total number of days in the crop growing season; T threshold The high temperature threshold is determined by historical baseline observation data, T. obs,hist The 95th quantile of the same sequence; f(x) is an indicator function that takes the value 1 when the condition is met and the continuous duration is ≥5 days, otherwise it takes the value 0.

[0063] Based on the daily precipitation P of the output fut,y Define extreme precipitation (R95p). y The total amount of precipitation exceeding the historical precipitation threshold is calculated using the following formula.

[0064] In the formula, P fut,y,d P represents the simulated daily precipitation value on day d of year y; 95 The value is the 95th quantile of daily precipitation ≥1 mm in the historical baseline daily precipitation series; D represents the total number of days. This indicator is used to characterize the potential for increased nitrogen and phosphorus loss from farmland and indirect greenhouse gas emissions G under future scenarios. ind Increased risk of heavy rainfall.

[0065] Based on P fut,y The consecutive drought days (CDD) are defined as the maximum number of consecutive days in a year with daily precipitation less than 1 mm. y ).

[0066] In the formula, This indicates the number of days with daily precipitation less than 1 mm.

[0067] When the number of drought days increases, on the one hand, it will affect crop yield through water stress, and on the other hand, due to the reduction of effective precipitation Pe, it will lead to an increase in irrigation water demand BW, thereby increasing irrigation energy consumption and emissions.

[0068] The identified extreme events are input into subsequent crop physiological models to capture the nonlinear damage of extreme high temperatures to crop pollen viability and grain filling rate, or the process by which water stress leads to stomatal closure or even permanent wilting.

[0069] In some embodiments, the digital twin model can be simulated using two driving forces: radiation-driven simulation and moisture-driven simulation. Specifically: The driving formula used in radiation-driven simulation is: In the formula, B represents the total dry matter biomass accumulated by the crop throughout its growth period; RUE represents the crop's light energy use efficiency; IPAR represents the total photosynthetically active radiation intercepted by the crop population; f(T), f(CO2), and f(W) represent the correction functions for the effects of temperature change, carbon dioxide concentration fertilizer effect, and water stress on crop growth, respectively. Simultaneously, the model outputs daily actual evapotranspiration for subsequent water requirement calculations.

[0070] The driving formula for water-driven simulation is: Where WP* represents the normalized crop water productivity parameter; ET c,i ET represents the daily actual evapotranspiration of the crop on day i; 0,i The reference crop evapotranspiration for the corresponding date is typically calculated using the FAO-56 Penman-Monteith equation. This approach focuses on characterizing the nonlinear limiting effect of water stress on yield.

[0071] The acquired records of actual management measures (such as irrigation volume and fertilization plans) and actual soil change data are input into the crop growth model in the digital twin environment. The model corrects the potential evapotranspiration based on the crop coefficient dynamically determined during the crop growth period and the water stress correction coefficient calculated based on soil moisture dynamics, thereby outputting the daily actual evapotranspiration ET, which reflects the true water consumption status of the farmland. c,i In some other embodiments, the process of predicting greenhouse gas emissions is explained below with a specific example: Based on the biomass production B and daily actual evapotranspiration ET determined above c,i Perform greenhouse gas emission calculations: In the formula, G rice Indicates methane emissions from rice paddies; G land Indicates nitrous oxide emissions from farmland; G mac Indicates energy consumption and emissions from agricultural machinery; G irri Indicates irrigation energy consumption emissions; G ind This indicates indirect greenhouse gas emissions.

[0072] Specifically: (1) Calculate the methane emissions (G) from paddy fields according to the following formula. rice ): In the formula, A rice Indicates the area planted with rice; EF rice The value represents the methane emission factor from paddy fields; 28 represents the global warming potential of methane relative to CO2 (GWP100).

[0073] (2) Calculate the nitrous oxide emissions from farmland (G) according to the following formula. land ): In the formula, N fert and N straw These represent the total nitrogen introduced by chemical fertilizers and straw return to the field; N indi For indirect nitrogen input from leaching and runoff; EF N2O 273 represents the emission factor; 273 represents the GWP100 of nitrous oxide.

[0074] (3) Calculate the agricultural machinery energy consumption emissions G according to the following formula. mac : In the formula, The output-output ratio coefficient per unit of fuel consumption; This is the greenhouse gas emission factor for diesel fuel.

[0075] (4) Calculate irrigation energy consumption emissions (G) according to the following formula. irri ): In the formula, BW is the value derived from ET. c,i Calculated irrigation water demand; H is the groundwater head; The proportion of groundwater irrigation; For the overall efficiency of the water pump; This is the electricity emission factor.

[0076] The formula for calculating BW is as follows: In the formula, BW represents the blue water (irrigation) requirement during the crop's growth period; ET c,i 10 represents the daily actual evapotranspiration; Pe represents the effective precipitation, calculated using the empirical formula recommended by the USDA Soil Conservation Bureau; 10 is the coefficient for converting depth to volume.

[0077] (5) Calculate indirect greenhouse gas emissions (G) using the following formula. ind ): In the formula, G ind This represents the total indirect greenhouse gas emissions generated by the crop production system in the upstream stage of its entire life cycle (i.e., before it enters the farmland); This category, categorized as agricultural inputs, covers all major agricultural production materials required to maintain simulated yield targets. It typically includes different types of fertilizers (nitrogen, phosphorus, and potash), pesticides, mulch film, and other agricultural inputs. jEF represents the physical consumption of the j-th agricultural input. This parameter is derived by back-calculating the biomass production demand predicted by the DSSAT or AquaCrop model, combined with historical resource input-output efficiency and future technological progress scenarios; j This represents the greenhouse gas emission factor of the j-th agricultural input during its industrial production, processing, packaging, and transportation from the production site to the farmland.

[0078] By using the aforementioned greenhouse gas emission factors, greenhouse gas emissions can be accurately predicted, enabling precise monitoring of carbon emissions.

[0079] Subsequently, the virtual yield sequence is compared with the actual yield in the original observation data to calculate the intervention effect value, which represents the degree of difference in yield before and after the intervention. This comparison is the core of counterfactual reasoning; by comparing the difference in crop yield under intervention conditions (virtual yield) and non-intervention conditions (actual yield), the true causal effect of the candidate factor on crop yield can be quantified. The intervention effect value can be the absolute difference, relative difference, or a standardized indicator between virtual and actual yields.

[0080] Finally, candidate factors with intervention effect values ​​greater than a preset threshold are identified as direct environmental factors with direct causal effects. The preset threshold is an empirical value or a critical value determined by statistical methods, used to judge the significance of the intervention effect. Only when the intervention effect value of a candidate factor exceeds this threshold is the factor considered to have a sufficiently significant direct causal effect on crop yield, thus being identified as a true direct environmental factor.

[0081] This application's embodiments, by introducing a digital twin environment and counterfactual intervention simulation, rigorously verify the causal effects of initially screened candidate factors, thereby enabling more accurate identification of environmental factors that truly have a direct causal effect on crop yield. This makes the generated set of sensitive factors more accurate and reliable, avoiding false alarms or missed alarms caused by the identification of incorrect factors, significantly improving the accuracy and effectiveness of agricultural food-water-energy-carbon monitoring and early warning, and providing a solid technical foundation for precision agricultural management.

[0082] In other embodiments, constructing counterfactual intervention scenarios for candidate factors in a digital twin environment includes: To identify the natural variation patterns of candidate factors in the time series, a time series decomposition algorithm based on Gaussian process regression is used to split the time series data of candidate factors into trend terms, seasonal terms, and random disturbance terms. While retaining the trend and seasonal components, the random disturbance component is replaced with a preset intervention disturbance pattern to generate intervention sequences corresponding to candidate factors; The intervention sequence is used as a new value of the candidate factor in the counterfactual scenario and input into the digital twin model to obtain the counterfactual intervention scenario of the candidate factor.

[0083] In this embodiment, time series decomposition involves breaking down a time series data into several components, typically including a trend term, a seasonal term, and a random disturbance term. The trend term reflects the long-term growth or decline trend of the data; the seasonal term captures the recurring patterns in the data within a fixed period; and the random disturbance term represents random fluctuations or noise in the data that cannot be explained by the trend and seasonality. Using a Gaussian process regression-based time series decomposition algorithm can effectively handle nonlinear relationships and uncertainties. By learning the covariance structure in the data, a smooth and robust estimation of each component can be achieved. For example, radial basis function (RBF) kernels or periodic kernel functions can be selected to capture different time series characteristics, and the model parameters can be optimized by maximizing the marginal likelihood, thereby accurately separating the original time series data into the aforementioned three components.

[0084] By retaining the trend and seasonal terms of the original time series, the intervention scenario is ensured to conform to the natural evolution of the environmental factor on a macroscopic level, avoiding unrealistic and drastic changes caused by the intervention. Simultaneously, replacing the random disturbance term with a pre-defined intervention disturbance pattern allows the intervention to precisely target the random, non-periodic portion of the factor's changes. The pre-defined intervention disturbance pattern can be designed according to actual research needs or potential intervention strategies; for example, it can be set as a continuous high-value disturbance (simulating continuous stress), a continuous low-value disturbance (simulating continuous improvement), a transient pulse-like disturbance (simulating a sudden event), or a periodic strengthening / weakening disturbance. By recombinating the replaced random disturbance term with the original trend and seasonal terms, a complete intervention sequence that incorporates both natural evolution patterns and specific intervention effects can be generated.

[0085] A digital twin model is a virtual model capable of simulating the behavior of real-world systems. In this application, it simulates the growth, development, and yield formation processes of crops under different environmental conditions. Using the generated intervention sequence as input to a candidate factor in a counterfactual scenario means that the digital twin model will simulate the crop response based on this carefully designed sequence that conforms to natural laws and includes specific interventions. In this way, the digital twin model can generate a highly realistic and scientifically grounded counterfactual intervention scenario, thereby providing reliable simulation data for subsequent evaluation of the causal effect of the candidate factor on crop yield.

[0086] When constructing counterfactual intervention scenarios, instead of simply setting intervention values, the time-series data of candidate factors are first refined and decomposed to separate their inherent trend, seasonal, and random disturbance components. This decomposition allows interventions to precisely target the random disturbance component while preserving the natural long-term and periodic variation patterns of the factors. Therefore, the generated intervention sequences, while maintaining the intervention objectives, better conform to the evolution patterns of actual environmental factors, avoiding distortions in counterfactual scenarios caused by unreasonable interventions. Inputting these more realistic intervention sequences into a digital twin model for simulation yields more accurate and reliable crop yield response trajectories, significantly improving the accuracy of causal effect assessments and ensuring higher credibility of identified direct environmental factors and their causal effects, providing a solid foundation for subsequent monitoring, early warning, and management decisions.

[0087] In this embodiment, S104 may include: For each sensitive factor in the sensitive factor set, a quantile regression model is constructed with the sensitive factor as the independent variable and multiple conditional quantiles of crop yield as the dependent variable. The quantile regression model estimates the functional relationship between factor values ​​and yield quantiles at multiple quantiles. From the quantile regression results, the conditional quantile function at each quantile is extracted, and the function curve corresponding to each quantile is taken as the output-factor response relationship at the quantile. For a preset risk quantile, search on the function curve of the risk quantile for the factor value that causes the production condition quantile to fall below the preset risk threshold for the first time, and determine the factor value as the optimal early warning threshold.

[0088] In this embodiment, each sensitive factor in the sensitive factor set is an environmental factor identified by the causal inference engine that has a direct causal effect on crop yield, such as soil moisture, temperature, light intensity, and nutrient content. The quantile regression model aims to deeply understand how these sensitive factors affect the entire distribution of crop yield, not just its average value. Unlike traditional least squares regression (OLS), which only focuses on the conditional mean of the dependent variable, quantile regression can estimate the influence of independent variables on different quantiles of the dependent variable (e.g., 10%, 25%, 50%, 75%, 90%). Specifically, for each sensitive factor, its historical observations are used as the independent variable, and the corresponding historical crop yield data is used as the dependent variable. By selecting a series of quantiles (e.g., 0.05, 0.1, 0.25, 0.5, 0.75, 0.9, 0.95, etc.), the quantile regression model fits an independent regression function at each selected quantile, thereby revealing the influence pattern of the sensitive factors at different yield levels. For example, a sensitive factor may have a small impact when crop yields are high, but a significant impact when crop yields are low (i.e., the lower quantile of the yield distribution).

[0089] After training, the quantile regression model outputs a conditional quantile function for each specified quantile. This function describes how the value of the sensitivity factor at a specific quantile establishes a quantitative relationship with crop yield. For example, for the 0.1 quantile (i.e., the lower 10% of the yield distribution), a function f_0.1(x) is obtained, indicating that when the sensitivity factor is x, there is a 10% probability that the crop yield is lower than f_0.1(x). These functions can be visualized as a series of curves, each representing a specific yield quantile. These curves together form a complete response map of the sensitivity factor to the crop yield distribution, clearly showing the strength and direction of the sensitivity factor's influence at different yield levels (from low to high yield).

[0090] A preset risk quantile refers to a yield quantile set by the user or system according to actual needs, representing a specific risk level. For example, a 0.1 quantile might represent a crop yield in the lowest 10% risk scenario. A preset risk threshold is a specific yield value; when the crop yield falls below this value, it is considered a significant yield loss or risk event. After determining the risk quantile (e.g., 0.1) and its corresponding conditional quantile function curve, the system searches for the value of the sensitive factor along this curve, starting from the normal range of values. When the value of the sensitive factor reaches a certain point, its corresponding yield conditional quantile (i.e., the predicted yield at that quantile) falls below the preset risk threshold for the first time. At this point, the value of the sensitive factor is determined as the optimal warning threshold. For example, if the risk quantile is set to 0.1 and the risk threshold is 200 kg / mu, the system will find a soil moisture value on the function curve at the 0.1 quantile, such that when the soil moisture is below this value, there is a 10% probability that the crop yield will be below 200 kg / mu. This soil moisture value is the optimal warning threshold.

[0091] This application's embodiments, by constructing a quantile regression model, can comprehensively capture the impact of sensitive factors on the entire distribution of crop yield, not just the average level. This allows the system to specifically focus on low-yield risk areas, i.e., the lower quantiles of the yield distribution, thereby more accurately identifying the critical factor values ​​that lead to yield decline. By searching for factor values ​​at preset risk quantiles that cause the yield condition quantile to first fall below a preset risk threshold, this application can automatically and dynamically determine the optimal early warning threshold. This threshold setting method based on risk quantiles can effectively avoid missed or false alarms caused by improper threshold settings, ensuring the timeliness and accuracy of early warnings, thus providing agricultural production managers with a more reliable decision-making basis and effectively reducing the risk of yield loss caused by abnormal fluctuations in environmental factors.

[0092] In some other embodiments, S105 may include: Obtain the set of sensitive factors and the corresponding causal graph structure. The causal graph structure contains a complete causal path starting from each sensitive factor, passing through at least one intermediate node, and finally reaching the crop yield node. The intermediate nodes include soil process variables, crop physiological variables, and microbial activity variables. On each monitoring unit in the target area, the current value of all factors in the sensitive factor set is collected in real time, and the current value of each sensitive factor is compared with the corresponding optimal warning threshold to identify the candidate sensitive factor that exceeds the threshold for the first time. For each candidate sensitive factor, based on the causal graph structure, determine the causal path from the factor to the output node and all intermediate nodes on the path; Real-time data of each intermediate node is collected sequentially along the causal path, and the real-time data of each intermediate node is compared with the preset intermediate threshold corresponding to the real-time data to determine the effective warning factor. The effective warning factor includes the candidate sensitive factor where the real-time data of all intermediate nodes on the path exceeds the threshold corresponding to the real-time data. Based on the spatial distribution of effective early warning factors and the degree of exceeding the threshold, multi-level monitoring results are generated.

[0093] In this embodiment, the causal graph structure can be pre-constructed in the following way: based on historical observation data, causal discovery algorithms (such as PC algorithm, fast causal inference algorithm or causal discovery algorithm based on score search) are used to learn the causal relationship between variables; at the same time, a knowledge graph in the field of agronomy is integrated to constrain and correct the data-driven causal graph to ensure that the causal direction conforms to the physiological laws of crops. The final generated causal graph is a directed acyclic graph, where nodes represent monitorable variables and edges represent causal relationships between variables.

[0094] At each monitoring unit in the target area, the current values ​​of all factors in the sensitive factor set are collected in real time. The monitoring unit can be a grid unit divided according to a preset resolution (such as 30m×30m or 100m×100m), and a corresponding sensor array is deployed in each unit.

[0095] The current value of each sensitive factor is compared with its corresponding optimal warning threshold to identify candidate sensitive factors that exceed the threshold for the first time. The optimal warning threshold can be predetermined by a threshold search algorithm, which learns the influence boundary of each factor on output based on historical data.

[0096] For each candidate sensitive factor, based on the obtained causal graph structure, the causal path from the factor to the output node and all intermediate nodes on the path are determined. If multiple paths exist, all are included in the subsequent verification scope.

[0097] For example, for the candidate sensitive factor "soil salinity", the causal path may include: "increased soil salinity → increased rhizosphere osmotic pressure → decreased crop evapotranspiration rate → decreased leaf stomatal conductance → decreased photosynthetic rate → reduced biomass accumulation → decreased final yield". The intermediate nodes in the path include monitorable or estimable variables such as "rhizosphere osmotic pressure", "evapotranspiration rate", "stomatal conductance", "photosynthetic rate", and "biomass accumulation".

[0098] Following a defined causal path, real-time data from each intermediate node is collected sequentially. The collection method varies depending on the node type: for nodes that can be directly monitored by sensors (such as soil moisture and temperature), the real-time readings of the corresponding sensors are retrieved; for nodes that are difficult to monitor directly (such as photosynthetic rate), they can be estimated through remote sensing inversion models or process models.

[0099] The real-time data of each intermediate node is compared with its corresponding preset intermediate threshold. The intermediate threshold can be determined by the threshold search algorithm described above, but the search target is "the causal effect boundary of this node on its downstream nodes". Only when the real-time data of all intermediate nodes on the path exceeds their corresponding thresholds is the candidate sensitive factor confirmed as a valid warning factor.

[0100] If the real-time data of an intermediate node on the path does not exceed the threshold, it is determined that the current anomaly has not been transmitted downstream along the causal path, the risk is naturally mitigated, the warning is terminated, and it is not included in the final monitoring results.

[0101] When a sensitive factor exists on multiple causal paths, chain verification is performed on each path. If at least one path passes verification, the factor is confirmed as a valid early warning factor, and the system records the verified path information as the basis for subsequent decision-making.

[0102] Based on the spatial distribution and threshold exceedance levels of all confirmed effective early warning factors, multi-level monitoring results incorporating causal chain verification information are generated. These monitoring results include: the name, spatial location, and threshold exceedance level of the effective early warning factor; the verified causal path; the real-time status of each intermediate node along the path; and the expected yield loss range calculated based on the path propagation strength.

[0103] In other embodiments, threshold quantization can also be used for threshold division, specifically: The resource and environmental pressures per unit of food output under future scenarios are assessed by calculating water use intensity (WI), energy use intensity (EI), and carbon emission intensity (CI) per unit of output using the following formulas: The Comprehensive Stress Index (SPI) is calculated using the following formula: In the formula, z i These are standardized values ​​for indicators related to water, energy, food, and carbon. These are the weighting coefficients determined using a combined weighting method. A higher SPI indicates a heavier resource and environmental burden on the region.

[0104] Based on the SPI quantification results, a tiered early warning system is implemented by setting thresholds. Taking into account regional resource endowments and policy constraints, a deviation degree of the pressure index is defined. The basis for threshold division: In the formula, SPI base The system's overall pressure value for the base year. When When the value is ≤0.1 or negative, the system is at low risk; when 0.1 < When the value is ≤0.3, the system enters a medium-risk warning; when If the risk level is >0.3 or the decline in grain production (B) exceeds the historical low, it is considered high risk.

[0105] By assessing and providing early warnings about carbon emissions through the methods described above, we can provide tiered early warnings for the sustainability of future agriculture. This will provide a scientific basis and refined decision-making support for resource-constrained regions to find the optimal path to balance food security and low-carbon transformation under complex future climate scenarios.

[0106] In some other embodiments, after S105, the method may further include: The sensitive factors that trigger early warnings and their corresponding real-time values ​​from the monitoring results are input into the micro-control module deployed in the field to generate electrical stimulation intervention instructions; According to the instructions for electrical stimulation intervention, a specific frequency of current is applied to the rhizosphere region of the crop to interfere with the metabolic activity of soil microorganisms; After the intervention ended, intervention-related production big data in the intervention area was collected again, and the intervention-related production big data was compared with the production big data before the intervention to verify the causal relationship and obtain the verification results. If the verification results indicate a weakening of the causal relationship, the early warning effectiveness of the sensitive factor is confirmed, and the weight of the sensitive factor is increased.

[0107] In this embodiment, the micro-control module is deployed in the field to receive sensitive factors that trigger early warnings and their corresponding real-time values ​​from monitoring results, and convert them into specific electrical stimulation intervention commands. This module can be an intelligent hardware device integrating a microcontroller, a communication unit (such as a wireless communication module), a power management unit, and a signal generator. When an early warning message is received, the micro-control module dynamically generates an electrical stimulation signal command with specific parameters (such as frequency, waveform, pulse width, duration, etc.) based on a preset strategy or algorithm, combined with the type of sensitive factor, the degree of exceeding the threshold, and the real-time value. For example, if monitoring results show an abnormally high activity of a certain anaerobic microorganism in the soil, the micro-control module can generate an electrical stimulation command aimed at inhibiting this type of microorganism or promoting its antagonistic microorganisms. The electrical stimulation intervention command is generated by the micro-control module based on the received early warning information and is used to guide the application of specific parameters of a current at a specific frequency in the crop rhizosphere. This command typically includes key information such as the frequency, intensity, duration, and waveform of the current. For example, the command might specify the application of a pulse current with a frequency of X Hz, an intensity of Y mA, and a duration of Z minutes. These parameters can be set based on previous experimental data, bioelectric effect models, or machine learning models to ensure that the target soil microbial metabolic activity is precisely targeted and the expected intervention effect is achieved.

[0108] A flexible electrode array is pre-embedded in the rhizosphere region of the crop, and these electrodes are connected to a micro-regulation module. Upon receiving an electrical stimulation intervention command, the micro-regulation module drives the electrodes to apply a current with a specific frequency and intensity to the soil. This current can alter the electrochemical properties of the soil microenvironment, thereby affecting the cell membrane permeability, enzyme activity, and gene expression of soil microorganisms, thus regulating their metabolic activity, such as inhibiting the growth of harmful microorganisms or promoting the reproduction of beneficial microorganisms. After the intervention, the system re-collects various production big data from the intervention area, including biomass yield, soil microcurrent signals, and soil physicochemical properties, forming intervention production big data. Subsequently, using causal inference methods, the intervention production big data is compared and verified with the production big data before the intervention. This comparison aims to identify whether there is a clear causal relationship between the intervention and the improvement of crop yield or soil health, excluding the influence of other confounding factors, thereby obtaining verification results. When the verification results show that the negative causal relationship between the sensitive factor that originally caused the early warning and the decline in crop yield is significantly weakened after the intervention, this strongly proves the effectiveness of the sensitive factor as an early warning indicator. Based on this, the system automatically adjusts the weight of the sensitive factor in the overall early warning model, increasing its weight. This means that in future early warning assessments, this sensitive factor will receive higher priority or greater influence, enabling the early warning system to identify and respond to similar potential risks more accurately and promptly.

[0109] When monitoring results identify sensitive factors that trigger early warnings, the system can immediately generate and execute electrical stimulation intervention commands through micro-control modules deployed in the field. This precisely applies a specific frequency of current to the crop rhizosphere, directly intervening in the metabolic activity of soil microorganisms and effectively mitigating or eliminating the adverse effects of sensitive factors on crop yield. This real-time, targeted intervention transforms early warning from passive alerts to proactive management, significantly improving the level of refined management and risk response capabilities in agricultural production. Furthermore, by re-collecting intervention-related production big data after the intervention and comparing it with pre-intervention production big data for causal verification, this application can scientifically assess the actual effectiveness of the intervention measures and further verify the effectiveness of the sensitive factors that trigger early warnings. When the verification results show that the negative causal relationship between the sensitive factor and crop yield is weakened due to the intervention, the system can automatically confirm the early warning value of the sensitive factor and adjust its weight in the early warning model accordingly. This feedback learning mechanism based on actual intervention effects enables the early warning system to continuously self-calibrate and optimize, improving its adaptability to complex agricultural environmental changes and the accuracy of early warnings. Ultimately, this achieves intelligent and adaptive management of the grain-water-energy-carbon system, ensuring stable and high yields of energy crops.

[0110] Based on the agricultural food-water-energy-carbon monitoring and early warning method provided in the above embodiments, this application also provides specific implementation methods for the agricultural food-water-energy-carbon monitoring and early warning device. Please refer to the following embodiments.

[0111] First see Figure 2 The agricultural food-water-energy-carbon monitoring and early warning device 200 provided in this application embodiment may include: The acquisition module 201 is used to acquire production big data of the target area. The production big data includes actual biomass yield, actual meteorological time series, actual management measures records and actual soil change data. Module 202 is used to perform spatial registration based on production big data and corresponding units in a preset static basic spatial database, and to build a performance-environmental condition matching dataset. The preset static basic spatial database includes a suitable zoning map for energy crop planting and an initial planting layout plan. The input module 203 is used to input the performance-environmental condition pairing dataset into the causal inference engine, which automatically identifies direct environmental factors that have a direct causal effect on crop yield through the causal structure learning algorithm, and removes interference factors that only have a correlation relationship, generating a set of sensitive factors for triggering early warning. The identification module 204 is used to automatically identify the optimal early warning threshold for each key factor affecting crop yield based on the set of sensitive factors and using a threshold search algorithm. The monitoring module 205 is used to monitor energy crops in the target area based on a set of sensitive factors and the corresponding optimal early warning threshold, and to obtain monitoring results. The monitoring results are used to monitor the food-water-energy-carbon balance in the target area.

[0112] As an optional implementation, the acquisition module 201 can be specifically used for: By deploying flexible array graphene electrodes in the rhizosphere of crops, soil microcurrent time-series signals are continuously acquired at millisecond-level frequencies. The soil microcurrent time series signal is input into a pre-trained carbon flux fingerprint recognition neural network to identify whether there are characteristic electrical signal waveforms in the time series signal that correspond to the denitrification process, methane oxidation process or carbon fixation metabolism process. When a characteristic electrical signal waveform is identified, the early warning lead time for an impending abnormal fluctuation in soil microbial metabolic activity is determined based on the occurrence time and amplitude parameters of the characteristic electrical signal waveform. This early warning lead time is then used as a supplementary field to the time dimension of production big data and simultaneously stored in the performance-environmental condition paired dataset.

[0113] As an alternative implementation, the construction module 202 can be specifically used for: For each sample in the performance-environmental condition paired dataset, overlay the corresponding historical early warning records and actual disaster loss records according to the spatial location of the sample. Using historical early warning records and actual disaster loss records as monitoring signals, a counterfactual reasoning algorithm is used to generate a causal effect difference between "virtual output if no early warning occurred" and "actual output if an early warning occurred" for each sample. The difference in causal effects is used as a weighting factor to weight the samples in the performance-environment condition paired dataset, generating a causal balanced dataset for training the causal inference engine. The causal balance dataset was identified as a new performance-environment condition pairing dataset.

[0114] As an alternative implementation, the input module 203 may be specifically used for: The performance-environmental condition pairing dataset is input into the initial causal structure learning model, and the set of candidate direct environmental factors and the corresponding initial causal graph structure are output through the constraint-based causal discovery algorithm. For each candidate factor in the set of candidate direct environmental factors, a counterfactual intervention scenario for the candidate factor is constructed in the digital twin environment. The counterfactual intervention scenario includes forcibly setting the candidate factor to a preset intervention value at a preset time point, while keeping the factors other than the initial candidate factor unchanged. Running a crop growth model coupled with carbon and nitrogen cycle mechanisms in a digital twin environment simulates the crop yield response trajectory under counterfactual intervention scenarios, generating a virtual yield sequence after intervention; The virtual yield sequence is compared with the actual yield in the original observation data to calculate the intervention effect value, which represents the degree of difference in yield before and after the intervention. Candidate factors whose intervention effect value is greater than a preset threshold are identified as direct environmental factors with direct causal effects.

[0115] As an alternative implementation, the input module 203 is also specifically used for: To identify the natural variation patterns of candidate factors in the time series, a time series decomposition algorithm based on Gaussian process regression is used to split the time series data of candidate factors into trend terms, seasonal terms, and random disturbance terms. While retaining the trend and seasonal components, the random disturbance component is replaced with a preset intervention disturbance pattern to generate intervention sequences corresponding to candidate factors; The intervention sequence is used as a new value of the candidate factor in the counterfactual scenario and input into the digital twin model to obtain the counterfactual intervention scenario of the candidate factor.

[0116] As an alternative implementation, the identification module 204 can also be used for: For each sensitive factor in the sensitive factor set, a quantile regression model is constructed with the sensitive factor as the independent variable and multiple conditional quantiles of crop yield as the dependent variable. The quantile regression model estimates the functional relationship between factor values ​​and yield quantiles at multiple quantiles. From the quantile regression results, the conditional quantile function at each quantile is extracted, and the function curve corresponding to each quantile is taken as the output-factor response relationship at the quantile. For a preset risk quantile, search on the function curve of the risk quantile for the factor value that causes the production condition quantile to fall below the preset risk threshold for the first time, and determine the factor value as the optimal early warning threshold.

[0117] As an alternative implementation, the monitoring module 205 can also be used for: Obtain the set of sensitive factors and the corresponding causal graph structure. The causal graph structure contains a complete causal path starting from each sensitive factor, passing through at least one intermediate node, and finally reaching the crop yield node. The intermediate nodes include soil process variables, crop physiological variables, and microbial activity variables. On each monitoring unit in the target area, the current value of all factors in the sensitive factor set is collected in real time, and the current value of each sensitive factor is compared with the corresponding optimal warning threshold to identify the candidate sensitive factor that exceeds the threshold for the first time. For each candidate sensitive factor, based on the causal graph structure, determine the causal path from the factor to the output node and all intermediate nodes on the path; Real-time data of each intermediate node is collected sequentially along the causal path, and the real-time data of each intermediate node is compared with the preset intermediate threshold corresponding to the real-time data to determine the effective warning factor. The effective warning factor includes the candidate sensitive factor where the real-time data of all intermediate nodes on the path exceeds the threshold corresponding to the real-time data. Based on the spatial distribution of effective early warning factors and the degree of exceeding the threshold, multi-level monitoring results are generated.

[0118] As an alternative implementation, the monitoring module 205 can also be used for: The sensitive factors that trigger early warnings and their corresponding real-time values ​​from the monitoring results are input into the micro-control module deployed in the field to generate electrical stimulation intervention instructions; According to the instructions for electrical stimulation intervention, a specific frequency of current is applied to the rhizosphere region of the crop to interfere with the metabolic activity of soil microorganisms; After the intervention ended, intervention-related production big data in the intervention area was collected again, and the intervention-related production big data was compared with the production big data before the intervention to verify the causal relationship and obtain the verification results. If the verification results indicate a weakening of the causal relationship, the early warning effectiveness of the sensitive factor is confirmed, and the weight of the sensitive factor is increased.

[0119] Figure 3 A schematic diagram of the hardware structure of the electronic device provided in an embodiment of this application is shown.

[0120] An electronic device may include a processor 301 and a memory 302 storing computer program instructions.

[0121] Specifically, the processor 301 may include a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of this application.

[0122] Memory 302 may include mass storage for data or instructions. For example, and not limitingly, memory 302 may include a hard disk drive (HDD), floppy disk drive, flash memory, optical disk, magneto-optical disk, magnetic tape, or Universal Serial Bus (USB) drive, or a combination of two or more of these. In one instance, memory 302 may include removable or non-removable (or fixed) media, or memory 302 may be non-volatile solid-state memory. Memory 302 may be internal or external to the integrated gateway disaster recovery device.

[0123] In one instance, memory 302 may be read-only memory (ROM). In one instance, the ROM may be a mask-programmed ROM, a programmable ROM (PROM), an erasable PROM (EPROM), an electrically erasable PROM (EEPROM), an electrically rewritable ROM (EAROM), or flash memory, or a combination of two or more of these.

[0124] Memory 302 may include read-only memory (ROM), random access memory (RAM), disk storage media device, optical storage media device, flash memory device, electrical, optical, or other physical / tangible memory storage device. Therefore, typically, memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., memory devices) encoded with software including computer-executable instructions, and when the software is executed (e.g., by one or more processors), it is operable to perform the operations described with reference to the agricultural food-water-energy-carbon monitoring and early warning method according to the first aspect of this disclosure.

[0125] The processor 301 reads and executes computer program instructions stored in the memory 302 to achieve... Figure 1 An agricultural food-water-energy-carbon monitoring and early warning method is shown in the embodiment.

[0126] In one example, the electronic device may also include a communication interface 303 and a bus 304. For example, Figure 3 As shown, the processor 301, memory 302, and communication interface 303 are connected through bus 304 and complete communication with each other.

[0127] The communication interface 303 is mainly used to realize communication between various modules, devices, units and / or equipment in the embodiments of this application.

[0128] Bus 304 includes hardware, software, or both, that couples components of an electronic device together. For example, and not as a limitation, the bus may include an Accelerated Graphics Port (AGP) or other graphics bus, an Extended Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), a HyperTransport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an Infinite Bandwidth Interconnect, a Low Pin Count (LPC) bus, a memory bus, a Microchannel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local (VLB) bus, or other suitable buses, or combinations of two or more of these. Where appropriate, bus 304 may include one or more buses. Although specific buses are described and illustrated in embodiments of this application, this application contemplates any suitable bus or interconnect.

[0129] This electronic device can execute the agricultural food-water-energy-carbon monitoring and early warning method in the embodiments of this application, thereby achieving a combination of Figures 1-2 The agricultural food-water-energy-carbon monitoring and early warning method and device are described.

[0130] Furthermore, in conjunction with the agricultural food-water-energy-carbon monitoring and early warning methods in the above embodiments, this application embodiment can provide a computer storage medium for implementation. This computer storage medium stores computer program instructions; when these computer program instructions are executed by a processor, they implement any one of the agricultural food-water-energy-carbon monitoring and early warning methods in the above embodiments.

[0131] In an optional embodiment, in conjunction with the agricultural food-water-energy-carbon monitoring and early warning method in the above embodiments, this application embodiment can provide a computer program product to implement it. The instructions in the computer program product are executed by the processor of an electronic device, enabling the electronic device to implement any of the agricultural food-water-energy-carbon monitoring and early warning methods in the above embodiments.

[0132] It should be clarified that this application is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of this application is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of this application.

[0133] The functional blocks shown in the above block diagram can be implemented as hardware, software, firmware, or a combination thereof. When implemented in hardware, they can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of this application are programs or code segments used to perform the required tasks. Programs or code segments can be stored on a machine-readable medium or transmitted over a transmission medium or communication link via data signals carried on a carrier wave. "Machine-readable medium" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROM, flash memory, erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, fiber optic media, radio frequency (RF) links, etc. Code segments can be downloaded via computer networks such as the Internet, intranets, etc.

[0134] It should also be noted that the exemplary embodiments mentioned in this application describe methods or systems based on a series of steps or apparatus. However, this application is not limited to the order of the above steps; that is, the steps can be performed in the order mentioned in the embodiments, or in a different order, or several steps can be performed simultaneously.

[0135] The aspects of this disclosure have been described above with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It should be understood that each block in the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that these instructions, executable via the processor of the computer or other programmable data processing apparatus, enable the implementation of the functions / actions specified in one or more blocks of the flowchart illustrations and / or block diagrams. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor, or a field-programmable logic circuit. It is also understood that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can also be implemented by special-purpose hardware performing the specified functions or actions, or can be implemented by a combination of special-purpose hardware and computer instructions.

[0136] The above description is merely a specific implementation of this application. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, modules, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. It should be understood that the protection scope of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the protection scope of this application.

Claims

1. A method for monitoring and early warning of agricultural food-water-energy-carbon resources, characterized in that, include: Acquire production big data for the target area, including actual biomass yield, actual meteorological time series, actual management measures records, and actual soil change data; Spatial registration is performed based on the corresponding units in the production big data and the preset static basic spatial database to construct a performance-environmental condition matching dataset. The preset static basic spatial database includes a suitable zoning map for energy crop planting and an initial planting layout scheme. The performance-environmental condition paired dataset is input into the causal inference engine. The causal structure learning algorithm automatically identifies direct environmental factors that have a direct causal effect on crop yield and removes interfering factors that only have a correlation relationship, generating a set of sensitive factors for triggering early warning. Based on the set of sensitive factors, a threshold search algorithm is used to automatically identify the optimal early warning threshold for each key factor affecting crop yield. Based on the set of sensitive factors and the corresponding optimal early warning threshold, energy crops in the target area are monitored to obtain monitoring results, which are used to monitor the food-water-energy-carbon balance in the target area.

2. The method according to claim 1, characterized in that, The acquisition of production big data in the target area includes: By deploying flexible array graphene electrodes in the rhizosphere of crops, soil microcurrent time-series signals are continuously acquired at millisecond-level frequencies. The soil microcurrent time series signal is input into a pre-trained carbon flux fingerprint recognition neural network to identify whether there are characteristic electrical signal waveforms in the time series signal that correspond to the denitrification process, methane oxidation process or carbon fixation metabolism process. When the characteristic electrical signal waveform is identified, the early warning advance amount for an impending abnormal fluctuation in soil microbial metabolic activity is determined based on the occurrence time and amplitude parameters of the characteristic electrical signal waveform. The early warning advance amount is then used as a supplementary field for the time dimension of production big data and simultaneously stored in the performance-environmental condition paired dataset.

3. The method according to claim 1, characterized in that, After spatial registration is performed on corresponding units in the production big data and the preset static basic spatial database to construct a performance-environment condition paired dataset, the method further includes: For each sample in the performance-environmental condition paired dataset, the corresponding historical early warning records and actual disaster loss records are superimposed according to the spatial location of the sample. Using the historical early warning records and actual disaster loss records as monitoring signals, a counterfactual reasoning algorithm is employed to generate a causal effect difference between "virtual output if no early warning occurred" and "actual output if an early warning occurred" for each sample. The difference in causal effects is used as a weighting factor to perform weighted sampling on the samples in the performance-environment condition paired dataset, generating a causal balance dataset for training the causal inference engine. The aforementioned causal balance dataset was identified as a new performance-environment condition paired dataset.

4. The method according to claim 1, characterized in that, The process of inputting the performance-environmental condition paired dataset into the causal inference engine, and automatically identifying direct environmental factors with a direct causal effect on crop yield through a causal structure learning algorithm, includes: The performance-environmental condition pairing dataset is input into the initial causal structure learning model, and the set of candidate direct environmental factors and the corresponding initial causal graph structure are output through the constraint-based causal discovery algorithm. For each candidate factor in the set of candidate direct environmental factors, a counterfactual intervention scenario for the candidate factor is constructed in a digital twin environment. The counterfactual intervention scenario includes forcibly setting the candidate factor to a preset intervention value at a preset time point, while keeping the factors other than the candidate factor unchanged. A crop growth model coupled with carbon and nitrogen cycle mechanisms is run in the digital twin environment to simulate the crop yield response trajectory under the counterfactual intervention scenario and generate a virtual yield sequence after the intervention. The virtual yield sequence is compared with the actual yield in the original observation data to calculate the intervention effect value, which represents the degree of difference in yield before and after the intervention. Candidate factors whose intervention effect value is greater than a preset threshold are identified as direct environmental factors with direct causal effects.

5. The method according to claim 4, characterized in that, The counterfactual intervention scenario for constructing the candidate factors in a digital twin environment includes: Identify the natural variation patterns of the candidate factors in the time series, and use a time series decomposition algorithm based on Gaussian process regression to split the time series data of the candidate factors into trend terms, seasonal terms, and random disturbance terms; While retaining the trend and seasonal terms, the random disturbance term is replaced with a preset intervention disturbance pattern to generate the intervention sequence corresponding to the candidate factor; The intervention sequence is used as a new value of the candidate factor in the counterfactual scenario and input into the digital twin model to obtain the counterfactual intervention scenario of the candidate factor.

6. The method according to claim 1, characterized in that, The step of automatically identifying the optimal early warning threshold for each key factor's impact on crop yield based on the set of sensitive factors using a threshold search algorithm includes: For each sensitive factor in the set of sensitive factors, a quantile regression model is constructed with the sensitive factor as the independent variable and multiple conditional quantiles of crop yield as the dependent variable. The quantile regression model estimates the functional relationship between factor values ​​and yield quantiles at multiple quantiles. From the quantile regression results, the conditional quantile function at each quantile is extracted, and the function curve corresponding to each quantile is taken as the output-factor response relationship at the quantile. For a preset risk quantile, search on the function curve of the risk quantile for the factor value that causes the production condition quantile to fall below the preset risk threshold for the first time, and determine the factor value as the optimal early warning threshold.

7. The method according to claim 1, characterized in that, The monitoring of energy crops in the target area based on the set of sensitive factors and the corresponding optimal early warning threshold yields monitoring results, including: Obtain a set of sensitive factors and the corresponding causal graph structure. The causal graph structure contains a complete causal path starting from each sensitive factor, passing through at least one intermediate node, and finally reaching the crop yield node. The intermediate nodes include soil process variables, crop physiological variables, and microbial activity variables. On each monitoring unit in the target area, the current value of all factors in the set of sensitive factors is collected in real time, and the current value of each sensitive factor is compared with the corresponding optimal warning threshold to identify the candidate sensitive factors that exceed the threshold for the first time. For each candidate sensitive factor, based on the causal graph structure, determine the causal path from the factor to the output node and all intermediate nodes on the path; Real-time data of each intermediate node is collected sequentially along the causal path, and the real-time data of each intermediate node is compared with the preset intermediate threshold corresponding to the real-time data to determine effective early warning factors. The effective early warning factors include candidate sensitive factors in which the real-time data of all intermediate nodes on the path exceeds the threshold corresponding to the real-time data. Based on the spatial distribution and threshold exceedance of the effective early warning factors, multi-level monitoring results are generated.

8. The method according to claim 1 or 7, characterized in that, After monitoring energy crops in the target area based on the set of sensitive factors and the corresponding optimal early warning threshold, and obtaining the monitoring results, the method further includes: The sensitive factors that trigger the early warning in the monitoring results and their corresponding real-time values ​​are input into the micro-control module deployed in the field to generate an electrical stimulation intervention command. According to the electrical stimulation intervention instructions, a current of a specific frequency is applied to the rhizosphere region of the crop to intervene in the metabolic activity of soil microorganisms; After the intervention ended, intervention-related production big data in the intervention area was collected again, and the intervention-related production big data was compared with the production big data before the intervention to verify the causal relationship and obtain the verification results. If the verification results indicate a weakening of the causal relationship, the early warning effectiveness of the sensitive factor is confirmed, and the weight of the sensitive factor is increased.

9. An agricultural grain-water-energy-carbon monitoring and early warning device, characterized in that, The device includes: The acquisition module is used to acquire production big data of the target area, including actual biomass output, actual meteorological time series, actual management measures records, and actual soil change data. The construction module is used to perform spatial registration based on the corresponding units in the production big data and the preset static basic spatial database to construct a performance-environmental condition matching dataset. The preset static basic spatial database includes a suitable zoning map for energy crop planting and an initial planting layout scheme. The input module is used to input the performance-environmental condition pairing dataset into the causal inference engine, which automatically identifies direct environmental factors that have a direct causal effect on crop yield through a causal structure learning algorithm, and removes interfering factors that only have a correlation relationship, generating a set of sensitive factors for triggering early warning. The identification module is used to automatically identify the optimal early warning threshold for each key factor affecting crop yield based on the set of sensitive factors and using a threshold search algorithm. The monitoring module is used to monitor energy crops in the target area based on the set of sensitive factors and the corresponding optimal early warning threshold, and to obtain monitoring results. The monitoring results are used to monitor the food-water-energy-carbon balance in the target area.

10. An electronic device, characterized in that, The device includes: a processor and a memory storing computer program instructions; When the processor executes the computer program instructions, it implements the agricultural food-water-energy-carbon monitoring and early warning method as described in any one of claims 1-8.

Images